Microsoft's Art Gallery demonstrates how materials normally presented in traditional media can successfully migrate to electronic publishing
Dick Pountain
Until recently, I'd been distinctly unimpressed by the quality of CD-ROMs, and I was beginning to think that the good old printed book was safe for another century. All it took to change my mind was Microsoft's new Art Gallery CD-ROM, which reproduces all 2200 paintings from the National Gallery in London.
The quality of the illustrations in Art Gallery is excellent--even on a humble SVGA display--and its hypertext user interface enables you to browse and make connections in ways that an art book simply can't match. As well as indexing the whole collection by name, subject matter, time, and geography, Art Gallery employs in
genious animations to explain painting techniques, compositional devices, color schemes, and symbolism that the painters used. It is like visiting the gallery with a team of art historians at your elbow, which is not surprising since the National Gallery's experts were intimately involved in designing Art Gallery.
Virtual Visits
A CD-ROM is no substitute for standing before the original paintings, but it can prepare you so that you know what you're looking for when you do visit a gallery. The fact that I live in London and regularly visit the National Gallery probably colors my enthusiasm for Art Gallery a little, but Microsoft has plans to produce similar CD-ROMs for other great art collections, and I hope it will use the same excellent user interface for them all.
I look forward to the day when I can have all the world's great paintings at my fingertips in a single bookshelf of CD-ROMs. In addition, I look forward to the day when all publishers follow the lead of the National Gallery in exa
mining beforehand how material can best be transported and displayed in electronic format. This is the secret to successful electronic publishing.
The Micro Gallery
Microsoft's Art Gallery CD-ROM is directly derived from an interactive on-line system called Micro Gallery, which is available free of charge to the public in the National Gallery's new Sainsbury Wing in Trafalgar Square. This new building, the first major extension to the gallery in 50 years, holds the gallery's early Renaissance collection. The Micro Gallery, which the American Express Foundation paid for with a $1,000,000 sponsorship, is a long thin room just off the ground-floor entrance, which holds a network of 12 keyboard-less Macintosh workstations that are built flush to the wall and fitted with 19-inch touch-sensitive screens. Visitors have free access to the Micro Gallery to browse the gallery's collection. For a small fee, you can make black-and-white prints of particular paintings or print a personalized map showing the loca
tions of a desired list of paintings.
The National Gallery was founded in 1824 to hold the U.K.'s national collection of western European paintings. It contains about 2200 works from over 700 artists, dating from the thirteenth to the early twentieth centuries and includes many major works of Rembrandt, Rubens, Titian, and Poussin, as well as a fine early Renaissance collection.
It is not perhaps the collection you would have expected to pioneer computerization--some high-tech twentieth-century gallery like Paris's Pompidou Center or the Getty would seem superficially to be a more likely candidate. A precondition of the National Gallery's success was the fact that its collection, while of outstandingly high quality, is relatively small, and almost all its pictures are on public display rather than in storage. This meant that a complete interactive catalog was feasible using personal computer technology from the late 1980s; however, museums with much larger collections would have to choose betwee
n using expensive mainframe systems or cataloging only parts of their collections. Even so a forward-looking management and a generous sponsor were vital in realizing the opportunity.
The gallery appointed an internal project manager who, in turn, recruited an outside software house, Cognitive Applications of Brighton, U.K., to create the necessary software. Graduates of the AI department of the nearby University of Sussex started this firm in 1985 and acquired special expertise in Macintosh programming and in designing interactive user interfaces. A team of up to 10 people from the gallery and Cognitive Applications put in 23 worker-years of development time spread over two-and-a-half years.
The design team quickly decided that the most important goal would be a fast response time, which would allow untrained visitors to browse the catalog in a spontaneous and relaxed fashion. Quantified, this meant that each painting should be retrieved in 1 second or less. Given the limits of 1989 personal co
mputer technology, this decision more or less determined the technical approach. Analog video storage had already been ruled out on grounds of poor visual quality, but having chosen to go digital, there was then no affordable network technology that would support an acceptable client/server implementation (if starting today, you might consider an ATM [Asynchronous Transfer Mode] network over optic fiber). Optical disk storage was also too slow.
Only magnetic storage could handle the job, so Micro Gallery was designed around Mac IIfx workstations (the fastest Mac then available). Each workstation contained a copy of the entire picture database on its local 1.3-GB hard drive. An Appletalk LAN is used outside public hours to perform maintenance updates only; it is not for interactive retrieval. This distributed storage scheme has the added advantage of being very robust; when one workstation goes down, it does not affect the others.
For displays, the team chose 19-inch Radius touchscreen monitors w
ith an extra fine dot pitch of 82 dpi (compared to the normal Mac 72 dpi) and a refresh rate higher than 70 Hz. To further enhance the appearance of these screens, all the text is antialiased in software.
The Micro Gallery catalog is a large publication of some 4500 pages with 300,000 words of text and 12,000 color illustrations; therefore, fitting it all onto even a 1.3-GB disk required further radical design decisions. To achieve acceptable picture quality, you might assume that the illustrations would have to be in 24-bit true color, but Cognitive Applications made no such assumption. Instead, it experimented with display depths of 24, 16, and 8 bits and discovered that 24-bit images were so large as to require hardware-assisted compression, and even then, the required response times could not be met.
Finally, the team bravely decided to go with 8-bit images, employing special palette optimization techniques to achieve adequate image quality. The strategy was spectacularly successful, and eve
n technically knowledgeable people who visit the Micro Gallery have great trouble believing the displays are not 24-bit. Plumping for 8-bit images also paved the way for Microsoft to turn Micro Gallery into the Art Gallery CD-ROM; only a few high-end Mac users and fewer PC users have 24-bit display hardware, so using 8-bit format opened up Art Gallery for the rest of us.
The Macintosh and Windows versions of Art Gallery contain almost exactly the same software used in the Micro Gallery, reduced in resolution from 1152 by 882 pixels to 640 by 480 pixels, which the less-expensive Macs and SVGA PCs can handle, and, of course, a mouse replaces touchscreen controls. If you try to run Art Gallery on a video adapter that doesn't support 256 colors, it warns you that the results will not be acceptable but will try it anyway; thus, you can even run it on a monochrome PowerBook. The Macintosh version of Art Gallery retains the antialiased body text, but the PC version had to drop this feature, as Windows doesn't
properly support multibit fonts.
Acquiring the Data
The Micro Gallery project and National Gallery CD-ROM benefited from some ongoing work at the National Gallery. In this regard, they are excellent examples of how the new media can leverage off the old. The National Gallery's photographic department already had a rolling program for photographing the complete collection onto 10- by 8-inch color transparencies, and in 1986, the National Gallery had produced a written catalog using computer typesetting. Both of these projects helped the Micro Gallery get off to a flying start. A handful of the less important pictures have only been photographed in black and white and appear as such in the Micro Gallery and the Art Gallery. Also, some pictures had been photographed in less-than-ideal conditions (e.g., tungsten light), and the team used Adobe Photoshop to color-correct these images manually.
It took two years to scan all the transparencies on a Sharp JX-600, for which Cognitive Applications had
to write custom scanner drivers to achieve the desired results. A strategic decision was made to scan everything at 24 bits so that scanning could begin before the final display depth was decided, and thus the scans could be reused if a 24-bit version was needed in the future. This demanded vast storage space, which was met in the early days by Syquest cartridges and later by erasable optical drives.
Most pictures in the Micro/Art Gallery appear at three different sizes: first as an 80- by 80-pixel thumbnail used in menu selections, then in position on the page, and finally zoomed to full-screen height. Run-time software enlargement from a single 8-bit master degrades the image quality too much, so three separate 8-bit bit maps are dithered down from each 24-bit master and stored in the run-time system.
Cognitive Applications evaluated various image-compression schemes, including the fractal compression scheme from Iterated Systems that Microsoft uses in its Encarta. The conclusion was that dit
hering to 8 bits is in itself the most effective and fastest compression for this particular material. Fractal compression of the 24-bit scans achieved a 3-to-1 compression ratio (as did going from 24 to 8 bits), but fractal images took longer to decompress and fractal compression lost more fine detail from the paintings. RLE (run-length encoding) schemes were ineffective in further compressing the dithered 8-bit images, which look pretty much like random bit streams to an RLE algorithm.
Optimizing the Palette
The excellent picture quality in the Micro/Art Gallery is entirely dependent on the choice of an optimal color mapping. Both Macintosh and PC SVGA 8-bit video adapters support 256-color display modes in which a CLUT (color lookup table) or palette holds the 256 current display colors chosen from some much larger color space (18 bits or around 250,000 colors for SVGA adapters). In practice, both the Macintosh and Windows operating systems reserve 16 or so colors for GUI purposes so that only ar
ound 240 colors are really selectable. Thus, the challenge is to choose exactly those 240 colors that will represent a particular painting or group of paintings as effectively as possible.
This task was made a little easier due to the National Gallery's collection containing mostly oil paintings over 100 years old, which use a more limited gamut of colors than is used today or even in early twentieth-century paintings. Finding the best palette involved a mix of computer and manual methods. Cognitive Applications wrote a program (see the text box "The Palette Optimization Algorithm") that analyzed a particular 24-bit scan and extracted a 256-color palette that distributed the color content of the picture as evenly as possible in a mathematical sense. In 1990, this was still fairly esoteric stuff, but nowadays, commercial tools like Adobe Photoshop include this capability, and indeed Cognitive Applications itself now uses Adobe Photoshop and DeBabelizer from Equilibrium Technology to extract palettes.
For the first Micro Gallery prototypes, the team tried to use a separate customized palette for each picture, but the adoption of menu pages containing several thumbnails made this impractical. Then the team tried one palette per page, but this still caused unsightly and disconcerting screen artifacts when turning from one page to another, and a solution using a special transition palette of neutral colors was only partly effective. It became clear that a single palette would have to work for the whole collection (see the figure "One Palette Fits All").
All 2200 scans were run through the color-mapping program and combined to yield a single 240-color palette. The National Gallery curators then reviewed the paintings as they appeared in this palette and suggested manual adjustments to the final mapping. This was a necessary step, because while a computer program can achieve a mathematically optimum mapping, it doesn't take any account of the subject matter and meaning of the paintings or of human a
esthetic preferences. For example, given a Bellini Virgin and Child, the program might correctly allocate as many shades of blue for the Virgin's cloak as flesh tones for the faces. But a human expert regards the faces as more important than the cloak and would like to dedicate more palette space to flesh tones, to achieve better modeling of faces and limbs without visible tone steps. Similar arguments might be made for the greens in landscapes, to which the human eye is particularly sensitive.
The final palette derived for the Micro/Art Gallery is enriched in flesh, green, and brown tones and relatively depleted in primary colors. If you were to use the same procedure to scan a collection of twentieth-century abstract paintings by Kandinsky, Rothko, and other artists, you might arrive at a completely different palette.
The Art Book that Moves
The most significant advantage that a computer screen offers over an art book is the possibility of moving the images, and the Micro/Art Gallery projec
t exploits this advantage by using animation to illustrate features of the paintings; another reason to be thankful for those 8-bit images.
Many of the animations simply consist of "draining" color from the painting to a monochrome image and then selectively reintroducing color to highlight certain groups of figures and compositional blocks. Another trick is to superimpose text labels that explain the symbolism of various objects in a picture. To illustrate perspective theories, you can overlay converging lines and grids or planes that cut up the picture space.
Other animations explore the way pictures were produced (e.g., by superimposing an x-ray view over the paint to reveal earlier work underneath). A Bellini fresco is dissected into the successive layers of applied paint, while a portrait of the duke of Wellington sprouts the extra medals Goya added at the end of the Peninsular War. Perhaps the most ambitious animation is for Vermeer's Young Woman Standing at a Virginal. To illustrate the a
ccuracy of Vermeer's perspective (he may have used a camera obscura), Art Gallery constructs a 3-D model from the painting and then flies through three different views of the room.
Some important paintings (e.g., Titian's Bacchus and Ariadne or Poussin's Adoration of the Golden Calf) warrant three or more animations dealing with composition, narrative, and construction methods. One animation that makes me smile is for the Adoration of the Golden Calf. It demonstrates that Poussin cribbed a group of figures from one of his own earlier paintings by simply flying the group from the earlier picture to the later while flipping it 180 degrees horizontally.
Invitation to Users
According to Cognitive Applications' director, Ben Rubenstein, the guiding principle in the design of Micro Gallery's user interface was "to make it more like a book than a screen." Many people still have a phobia about computers, and people who regularly visit art galleries might be expected to be less at ease with computers
than the average. The result of following this principle is a roaring success judged from the overwhelmingly favorable comments in the National Gallery visitor's book, and the fact that utilization of the workstations runs at close to 100 percent every day.
The graphics style of the screens is very book-like, with black-on-white text set in a serif typeface arranged in columns around the pictures. The strict avoidance of colored backgrounds or lettering in the text enhances the impact of the paintings and makes them the focus of interest on the page. All computerish "screen furniture" is confined to a narrow navigation bar at the foot of the screen, and no icons, scroll bars, or menu bars are in sight. "Hot text" that invokes a hypertext link to another page or a pop-up information window is lightly shaded, and you quickly learn that clicking any shaded area will take you somewhere else. Click on any picture marked with the little four-arrows sign to zoom it and then click anywhere to unzoom it again.
Fingertip Control
Hypertext-links aside, navigation is performed using four buttons at the bottom of the screen. The two most important buttons are Next Page, which allows you to turn the page like a book, and Go Back, which allows you to return to the last page you saw (even if you left it via a hypertext jump) and retain an effectively unlimited memory trail of pages.
The Contents button takes you straight to the contents page, which is the nerve center of the browsing system. You are offered five different viewpoints into the National Gallery collection: The Painter's Lives view lets you look up painters by name; the Historical Atlas lets you look at a particular country and a particular century; Picture Types enables you to browse landscapes, still lives, portraits, nudes, and whatever; General Reference is an overall index so that you can find an explanation of, say, gesso or chiaroscura; and finally Guided Tours offers you one of four automated sessions with themes like Composition and
Perspective and Making paintings.
On the Art Gallery CD-ROM, a guide's high-quality digitized voice accompanies you on these tours; however, for the Micro Gallery application, sound was deemed too distracting to other users. In addition, Art Gallery uses voice to excellent effect in the Painter's Lives section to give the correct pronunciation of names. This feature is very welcome indeed when you are faced with a Pollaiuolo, Koekkoek, or Reymerswaele.
The See Also button presents you with a context-sensitive contents page that offers you cues for further browsing--painters of the same period or similar style or paintings whose subject matter is related to the one you are currently viewing. This is the most innovative feature of Art Gallery, and a few hours spent following the See Also cues can be a genuinely educational experience. A dreaded pop-up menu is also available for the experienced computer user. The menu is well hidden behind the Options button, which allows you to print or copy the s
creen to the clipboard and perform text searches on the entire CD-ROM.
It's no accident that the user interface to Art Gallery is much simpler and more intuitive than the Macintosh/Windows GUIs that support it, for Cognitive Applications tried out many prototypes of the system on museum staff and members of the public and then stripped away or refined features to arrive at the current model. Perhaps art history is not the only thing you can learn in this Art Gallery.
Figure: One Palette Fits All
The decision to use a single palette across the collection made choosing the palette decisive in the success of the publishing effort. After the computers were finished, experts from the National Gallery tweaked the palette so that it worked across the range of sty
les represented by the collection.
Landscape with Poplars, Cezanne
Combing the Hair, Degas
The Supper at Emmaus, Caravaggio
The Wilton Diptych, anonymous
Dick Pountain is a BYTE contributing editor based in London. You can reach him on the Internet or BIX at
dickp@bix.com
.
The Fine Art of CD-ROM Publishing
